Chen Weifu, Xu Jianrong, Chiu Bernard
Department of Electronic Engineering, City University of Hong Kong, Hong Kong.
Renji Hospital, Shanghai Jiao Tong University, Shanghai 200127, China.
Med Phys. 2015 May;42(5):2431-48. doi: 10.1118/1.4916803.
The peripheral arterial disease is a powerful indicator of coexistent generalized atherosclerosis. As plaques in femoral arteries are diffused and can span a length of 30 cm, a large coverage of the arteries is required to assess the full extent of atherosclerosis. Recent development of 3D black-blood magnetic resonance imaging sequences has allowed fast acquisition of images with an extended longitudinal coverage. Vessel wall volume quantification requires the segmentation of the lumen and outer wall boundaries, and conventional manual planimetry would be too time-consuming to be feasible for analyzing images with such a large coverage. To address this challenge in image analysis, this work introduces an efficient 3D algorithm to segment the lumen and outer wall boundaries for plaque and vessel wall quantification in the femoral artery.
To generate the initial lumen surface, a user identified the location of the lumen centers manually on a set of transverse images with a user-specified interslice distance (ISD). A number of geometric operators were introduced to automatically adjust the initial lumen surface based on pixel intensity and gradient along the boundary and at the center of each transverse slice. The adjusted surface was optimized by a 3D deformable model driven by the local stiffness force and external force based on image gradient. The optimized lumen surface was expanded to obtain the initial outer wall surface, which was subsequently optimized by the 3D deformable model.
The algorithm was executed with and without adjustment of the initial lumen surface and for three different selections of ISD: 10, 20, and 30 mm. The segmentation accuracy was improved in a statistically significant way with the introduction of initial lumen surface adjustment, but was insensitive to the ISD setting. When compared with the manual segmentation, the settings with adjustment have, on average, mean absolute differences (MADs) of 0.28 and 0.36 mm, respectively, for lumen and outer wall segmentations, which are significantly lower than those obtained when the adjustment operators were not applied (MAD = 0.43 and 0.59 mm for lumen and outer wall segmentations). The algorithm took about 1% of the time required for manual segmentation to complete segmenting the whole 3D femoral artery.
The proposed semiautomatic algorithm generated accurate lumen and outer wall boundaries from 3D black-blood MR images with few user interactions, thereby allowing rapid and streamlined assessment of plaque burden in the femoral arteries.
外周动脉疾病是并存的全身性动脉粥样硬化的有力指标。由于股动脉中的斑块呈弥漫性，长度可达30厘米，因此需要对动脉进行大面积覆盖，以评估动脉粥样硬化的全貌。三维黑血磁共振成像序列的最新发展使得能够快速采集具有扩展纵向覆盖范围的图像。血管壁容积定量需要对管腔和外壁边界进行分割，而传统的手工平面测量法对于分析如此大面积的图像来说太耗时，难以实施。为应对图像分析中的这一挑战，本研究引入了一种高效的三维算法，用于分割股动脉中的管腔和外壁边界，以进行斑块和血管壁定量分析。
为生成初始管腔表面，用户在一组具有用户指定层间距（ISD）的横向图像上手动确定管腔中心的位置。引入了一些几何算子，根据每个横向切片边界处和中心处的像素强度及梯度自动调整初始管腔表面。通过基于图像梯度的局部刚度力和外力驱动的三维可变形模型对调整后的表面进行优化。将优化后的管腔表面扩展以获得初始外壁表面，随后通过三维可变形模型对其进行优化。
该算法在有和没有调整初始管腔表面的情况下执行，并且针对三种不同的ISD选择：10、20和30毫米。引入初始管腔表面调整后，分割精度有统计学意义的提高，但对ISD设置不敏感。与手工分割相比，调整设置的情况下，管腔和外壁分割的平均绝对差（MAD）分别为0.28和0.36毫米，显著低于未应用调整算子时获得的结果（管腔和外壁分割的MAD分别为0.43和0.59毫米）。该算法完成整个三维股动脉分割所需的时间约为手工分割所需时间的1%。
所提出的半自动算法通过很少的用户交互从三维黑血磁共振图像中生成了准确的管腔和外壁边界，从而能够快速、简化地评估股动脉中的斑块负荷。
